Conclusion

The method described here uses artificial intelligence tools to design new alloys. In particular, it saves costs compared with a trial-and-error method, since it uses numerical calculation, which is less costly than experimentation. It also saves development time. These remain modest, however, as the method requires a great deal of work: setting up and "cleaning" databases, testing and validating various configurations of mining models, parameterizing optimization algorithms (defining and testing different configurations of objectives and constraints, seeding) and final experimental validation. The latter sometimes highlights the need to return to the design phase to adjust the parameterization. On the other hand, combining optimization using surrogate models (§